entropy

Article
Training Multilayer Perceptron with Genetic
Algorithms and Particle Swarm Optimization for
Modeling Stock Price Index Prediction
Fatih Ecer 1 , Sina Ardabili 2,3 , Shahab S. Band 4 and Amir Mosavi 5,6,7, *
1 Department of Business Administration, Afyon Kocatepe University, Afyonkarahisar 03030, Turkey;
fecer@aku.edu.tr
2 Biosystem Engineering Department, University of Mohaghegh Ardabili, Ardabil 5619911367, Iran;
s.ardabili@ieee.org
3 Kando Kalman Faculty of Electrical Engineering, Obuda University, 1034 Budapest, Hungary
4 Future Technology Research Center, College of Future, National Yunlin University of Science and Technology,
123 University Road, Section 3, Douliou, Yunlin 64002, Taiwan; shamshirbands@yuntech.edu.tw
5 Faculty of Civil Engineering, Technische Universität Dresden, 01069 Dresden, Germany
6 Institute of Research and Development, Duy Tan University, Da Nang 550000, Vietnam
7 School of Economics and Business, Norwegian University of Life Sciences, 1430 As, Norway
* Correspondence: amir.mosavi@mailbox.tu-dresden.de or amirhoseinmosavi@duytan.edu.vn




Received: 22 August 2020; Accepted: 28 October 2020; Published: 31 October 2020


Abstract: Predicting stock market (SM) trends is an issue of great interest among researchers, investors
and traders since the successful prediction of SMs’ direction may promise various benefits. Because
of the fairly nonlinear nature of the historical data, accurate estimation of the SM direction is a
rather challenging issue. The aim of this study is to present a novel machine learning (ML) model to
forecast the movement of the Borsa Istanbul (BIST) 100 index. Modeling was performed by multilayer
perceptron–genetic algorithms (MLP–GA) and multilayer perceptron–particle swarm optimization
(MLP–PSO) in two scenarios considering Tanh (x) and the default Gaussian function as the output
function. The historical financial time series data utilized in this research is from 1996 to 2020,
consisting of nine technical indicators. Results are assessed using Root Mean Square Error (RMSE),
Mean Absolute Percentage Error (MAPE) and correlation coefficient values to compare the accuracy
and performance of the developed models. Based on the results, the involvement of the Tanh (x) as
the output function, improved the accuracy of models compared with the default Gaussian function,
significantly. MLP–PSO with population size 125, followed by MLP–GA with population size 50,
provided higher accuracy for testing, reporting RMSE of 0.732583 and 0.733063, MAPE of 28.16%,
29.09% and correlation coefficient of 0.694 and 0.695, respectively. According to the results, using the
hybrid ML method could successfully improve the prediction accuracy.

Keywords: stock market; machine learning; multilayer perceptron; financial data; artificial
intelligence; artificial neural networks; online trading; big data; social science data; evolutionary
algorithms; optimization

1. Introduction
Accurately predicting the stock market (SM) index direction has frequently been a topic of great
interest for many researchers, economists, traders and financial analysts [1]. Nonetheless, the SM field
is neither static nor predictable. In fact, SM trends are sensitive to both external and internal drivers.
Thus, SM index movement estimation can be categorized under complex systems [2]. Stock price
movement is often interpreted as the direction of stock price and used for prediction. Determining the

Entropy 2020, 22, 1239; doi:10.3390/e22111239 www.mdpi.com/journal/entropy

Entropy 2020, 22, 1239 2 of 19

future direction of stock price movement is of utmost importance to investors to evaluate the market
risks. Predicting direction of stock price movement has been seen as a challenging and complex task to
model [3]. Complex system is a framework to work how a system’s sub-categories interact with each
other and how the whole system interacts and manages relationships with its environment. Whereas
modeling such complex systems, challenges have encountered during constructing a reliable and
effective technique and deciding its architecture [4].
Stock price movement forecasting is a compelling task because of the high volatility, anomaly
and noisy signal in the SMs’ area. Over the past two decades, this topic has attracted the attention of
researchers in different fields, particularly artificial intelligence [5]. Stock prices are nonlinear with
regard to historical data and other technical and macroeconomic indicators [6]. Many researchers often
preferred to use time-series analyses which is utilized to estimate future events according to historical
data before the capabilities of neural networks were discovered. Autoregressive integrated moving
average (ARIMA), autoregressive conditional heteroskedasticity (ARCH) model and Generalized
autoregressive conditional heteroskedasticity (GARCH) model, support vector machine (SVM) are
among the best-known models among the methodologies [7]. Moreover, regression analysis and
artificial neural networks (ANNs) have frequently been used for forecasting and classification in
order to cope with these nonlinear relationships [7–10]. Systems that utilize technical analysis,
across expert systems, hybrid systems and various types of computational intelligence have as well
been suggested [11–13]. Interests of researchers continue to increase for applying different types of
artificial intelligence to forecast SM index direction. Due to the nonlinear structures of the problems,
the prediction approaches are typically highly complex, meanly needing to develop efficient solution
methods for such models. Technical analysts work with many data, technical tools and especially
technical indicators to decide price trends and market trends on the basis of price and volume
conversions in the market [14]. In light of existing literature, several technical indicators have been
preferred as input data in the creating of forecasting methodologies to predict the direction of a
SM index [15–22]. Cervelló Royo and Guijarro, 2019 employed four ML based prediction methods
including gradient boosting machines (GBM), random forest (RF), generalized linear models (GLM)
and deep learning (DL) for to address the estimation of market trends as a comparison analysis by
accuracy rate (%) [23].
In several studies, technical indicators are used as input data. The trend was to identify novel
methods that provide the highest accuracy for the classification methods in the index direction
prediction. When we provide real-valued technical indicators as inputs to the models, forecasting
techniques to be classified in accordance with the values of technical indicators are created [15,20].
When technical indicators are utilized, the prediction models consider each indicator as input data,
regardless of the index being considered [7]. The studies in which the methods that deal with technical
indicators as input data are used in the index direction prediction constitute the framework of this study.
Estimating the BIST 100 index’s direction is a significant financial issue that has carefully monitored in
financial markets around the world [24,25]. In this context, this research aims to predict stock price
movement direction through an integrated multilayer perceptron methodology. More specifically,
two novel models, i.e., multilayer perceptron–genetic algorithms (MLP–GA) and multilayer–particle
swarm optimization (MLP–PSO) with Tanh (x) as the output function, have been proposed and applied
for prediction compared with the default Gaussian function. Thus, it is intended to fill a gap in SM
direction prediction literature. PSO has been employed by researchers for the prediction of stock
market. Lahmiri, 2018 developed spectrum analysis and support vector regression integrated with
PSO to estimate the stock price using time series data [26]. Pulido et al., 2014 employed PSO for
the optimization of a hybrid ANN–Fuzzy model for the prediction of Mexican Stock Exchange [27].
Lahmiri, 2016 employed PSO for the optimization of the architecture of a feed forward neural network
in the prediction of stock market [28].
Entropy 2020, 22, 1239 3 of 19

The rest of this research is organized as follows: The next section reviews the relevant literature.
Section 3 deals with the research methodology. The results are given in Section 4, while the findings
are discussed in Section 5. The conclusions are presented in the final section.

2. Literature Review
In recent years, there have been a great deal of papers investigating the direction of the next day
trends of SMs. Academicians and traders have made enormous efforts to forecast the next day trends
of SM index for translating the predictions into profits (Kara et al., 2011). In this section, we focus the
review of methods and technical indicators utilized for forecasting of direction movement of stock
index. As shown in Table 1, an ANN model was used in some of the studies [18,25,29,30], whilst hybrid
models were preferred in other studies [17,21,31–33] as displayed in Table 2. In Table 1. the notable
algorithms are back-propagation neural network (BPNN), independent component analysis- BPNN
(ICA–BPNN), Naive Bayes (NB), and k-nearest neighbors algorithm (k-NN).
Table 1 summarizes several machine learning methods proposed for stock exchange index
direction prediction. The most popular methods are RF, SVM and ANN, followed by k-NN and NB,
with dissimilar accuracy outcomes. The state of the art shows a research gap in using hybrid models.
Table 2 summarizes the notable machine learning models. Hybrid models of the SVM trained
with simple evolutionary algorithms such as GA have been the most popular. The state of the art of
hybrid models shows a research gap in using more sophisticated machine learning models trained
with advanced soft computing techniques.

Table 1. Stock market index direction forecasting with machine learning considering comparative
analysis involving ANN-based methods.

References Method/s Application/Data Result

[15] ANN, ARIMA KLCI (1984–1991) ANN outperformed ARIMA model.
[16] SVM, BPNN KOSPI (1989–1998) SVM outperformed ANN.
[32] ICA–BPNN, BPNN TAIEX (2003–2006) ICA–BPNN is superior.
[17] ANN, NB, DT BSE (2003–2010) Hybrid RSs outperformed ANN.
[34] PNN, SVM S&P 500 (2000–2008) PNN provided high accuracy.
[24] ANN, SVM BIST 100 (1997–2007) 75% accuracy using ANN.
[29] ANN TEPIX (2002–2009) ANN showed promising results.
[35] ANN, GA TEPIX (2000–2008) ANN delivered next day estimates.
[25] ANN BIST 100 (2002–2007) ANN achieved success with 82.7%.
[36] SVM, ANN IBEX-35 (1990–2010) SVM outperformed ANN.
[37] k-NN, PNN S&P 500 (2003–2008) k-NN outperformed PNN.
[18] ANN BOVESPA (2000–2011) ANN suitable for direction estimation.
[38] LSSVM, PNN, CSI 300 (2005–2012) LSSVM outperformed other models.
Random walk, ANN, The fuzzy metagraph-based model has reached
[39] BSE-SENSEX (2011–2012)
SVM, fuzzy a classification rate of 75%.
[40] ANN, RF, k-NN Amadeus (2009–2010) RF outperformed ANN.
[20] NB, ANN, SVM CNX Nifty (2003–2012) NB outperformed other models.
[41] DWT, ANN, SVM–MLP DJIA-S&P500 (2000–2012) SVM–MLP is superior.
[42] Probit, Logit, Extreme Value S&P 500 (2011–2015) Extreme Value outperfomed Logit and Probit.
[7] PSO–ANN S&P 500, IXIC (2008–2010) Acceptable prediction and robustness.
[43] RF & ANN S&P 500 (2009–2017) RF outperformed ANN.
[44] Hybrid fuzzy NN DAX-30 (1999–2017) minimum risky strategies.
BM&FBOVESPA
[45] GA, SVM, ANN SVM performed better than ANN.
PETR4 (1999–2017)
Entropy 2020, 22, 1239 4 of 19

Table 2. Stock market index direction studies using methods other than ANNs.

Author/s Method/s Application Result

GA–SVM, random walk, SVM, S&P 500 GA–SVM has been shown to outperform
[31]
ARIMA, BPNN (2000–2004) other models.
Fuzzy sets, physical, support
TAIEX and HIS Their proposed models outperform the
[14] vector regression, partial least
(1998–2006) compared models according to the RMSE.
squares regression
CROBEX Random forests can be successfully
[46] Random forest
(2008–2013) preferred to estimate.
Apple company The fuzzy expert system has significant
[19] Fuzzy rule-based expert system
(2010–2014) performance with minimal error.
Indonesia ASII.JK The GMM–SVM model has been found to
[21] GMM–SVM
(2000–2017) be superior to other models.
iBOVESPA Mean accuracy with the proposed model
[47] Bayesian network
(2005–2012) configuration was almost 71%.
While SVM model performs better in BSE
TOPSIS, SVM, NB, BSE SENSEX, S&P500
[48] SENSEX index, k-NN is superior to other
Decision tree, kNN (2015–2017)
models in S&P 500 index.
Apple stock data The proposed method outperformed the
[22] ANFIS
(2005–2015) existing methods.
BSE, HIS, FTSE They proved the superiority of the RKELM
[49] RKELM
(2010–2015) model over the ANN, naive Bayes and SVM.
DJIA, S&P500, HSI,
[3] Mean Profit Rate (MPR) Nikkei 225, SSE MPR is an effective classifier.
(2007–2017)

Technical Indicators
As mentioned above, technical indicators have been useful and effective financial instruments
for estimating direction of stock price index for years. Technical indicators used for SM direction
prediction from past to present can be seen in Table 3.
In summary, MACD, %K, %D, RSI, %R, A/D, MOM, EMA, CCI, OSCP and SMA are the technical
indicators that are frequently preferred by researchers. Furthermore, ANN and its extensions (MLP,
PNN, etc.) are the most used methods. As far as the authors know, MLP–GA and MLP–PSO
methodologies with and without Tanh (x) as the output function have not been proposed to forecast
stock exchange movement prediction for any stock exchange in the literature. As a result, it is
anticipated that this paper will constitute a significant contribution to the related field.

Table 3. Technical indicators used in SM direction estimation.

Author/s Technical Indicators

Simple moving average (SMA), stochastic K (%K), momentum (MOM), stochastic D (%D),
[15]
relative strength index (RSI).
Slow D%, MOM, rate of change (ROC), K%, Larry William’s R% (%R),
[16] Accumulation/Distribution (A/D) oscillator, disparity5, RSI, disparity10, price oscillator
(OSCP), D%, Commodity Channel Index (CCI).
OSCP, Stochastic oscillator (SO), Slow stochastic oscillator (SSO), CCI, ROC, MOM, SMA,
Moving variance (MV), Moving variance ratio (MVR), Exponential moving average (EMA),
[31]
Moving average convergence and divergence (MACD), A/D oscillator, Price (P), disparity5,
disparity10, Moving stochastic oscillator (MSO), RSI, linear regression line (LRL).
The previous day’s cash market high, low, volume, 6-day RSI, today’s opening cash index,
[32]
10-day total amount weighted stock price index.
%K, Positive volume index, %R, negative volume index, %D, on balance volume, RSI,
[17]
MACD, MOM, A/D oscillator, 25-day SMA.
[34] SMA, OSCP, MOM, %D, ROC, disparity, %K.
MACD, SMA, %R, CCI, A/D oscillator, %D, weighted moving average (WMA), RSI,
[29]
MOM, %K.
[24] %D, %K, RSI, MOM, MACD, WMA, %R, A/D oscillator, SMA, CCI.
[35] SMA, MACD, RSI, OSCP, MOM, volume.
Entropy 2020, 22, 1239 5 of 19

Table 3. Cont.

Author/s Technical Indicators

[18] MACD, RSI, %D, SMA, Bollinger band, MOM, %R.
SMA for 5 days, SMA for 10 days, bias to moving average (BIAS), RSI, psychological line
[14]
(PSY), %R, MACD, MOM.
[38] %K, %R, %D, CCI, A/D oscillator, MOM, MACD, RSI, SMA and WMA.
[39] MA, exponential moving average (EMA), MACD, RSI.
[19] High price, low price, volume, change of closed price, MACD, MA, BIAS, RSI, %R.
[20] %D, RSI, WMA, MACD, CCI, A/D oscillator, %K, %R, SMA.
5-day SMA, 5-day WMA, 10-day SMA, 10-day WMA, %K, %D, MACD, CCI, 5-day disparity,
[46]
10-day disparity, OSCP, ROC, MOM, RSI, 5-day standard deviation.
[41] SMA, EMA, A/D oscillator, %K, RSI, OSCP, closing price, maximum price.
[42] SMA, WMA, MOM, %K, %D, %R, RSI, MACD.
Change of price, change of volume, 5-day SMA, 10-day SMA, 30-day SMA, moving price
[7]
level (30 days), moving price level (120 days), percentage price oscillator.
A/D oscillator, mean of rising days, CCI, SMA, MACD, MOM, on balance volume, ratio of
[21]
rising days, RSI, %R.
Triangular moving average (TMA), RSI, SMA, EMA, modified moving averages (MMA),
[44]
volatility ratio (VR), %R, true strength index (TSI), average true range (ATR).
[48] SMA, %K, %D, %R, MACD, RSI.
[45] SMA, WMA, MOM, RSI.
[22] 1-week SMA, 2-week SMA, 14-day disparity, R%.
[3] %D, %K, RSI, MOM, MACD, WMA, %R, A/D oscillator, SMA, CCI.
[49] SMA, MACD, %K, %D, RSI, %R.

3. Materials and Methods

3.1. Data
As shown in Table 4, nine technical indicators for each trading day were utilized as input data.
Plenty of investors and traders handle certain criteria for technical indicators. A great deal of technical
indicators is available. As already mentioned above, technical indicators have often been considered
as input variables in the construction of forecasting systems for estimating the trend of movement
of SM index [24]. As a result, we determined nine technical indicators by previous studies and the
opinion of area experts.

Table 4. Selected technical indicators.

Technical Indicators Abbreviation Formulas

Simple n (10 here)-day Moving Average SMA SMA = Ct + Ct − 1n+···+ Ct − n
Simple n (10 here)-day Moving Average WMA WMA = 10 ×n C+t (+n 9−×1)C+
t−1 +···+C1
···+ 1
Momentum MOM MOM = Ct − Ct−9
Pn−1
Stochastic D% STOCH i=0 Kt−i
Stokastik%D = 10 %
100 P
Relative Strength Index RSI RSI = 100 −
1+( i=0 UPt−i /n)/( n−1
Pn−1
i=0 DWt−i /n)
Moving Average Convergence 2
MACD MACD = MACD(n)t−1 + n+1 × (DIFFt − MACD(n)t − 1 )
Divergence
Hn − Ct
Larry William’s R% LWR William0 s%R = Hn − Ln × 100
Ht − Ct − 1
Accumulation/Distribution Oscillator A/D A/D = Ht − Lt
Mt − SMt
Commodity Channel Index CCI CCI = 0.015Dt

Ct is the closing price, Lt the low price, Ht the high price at time t, DIFF: EMA(12)t – EMA(26)t , EMA exponential
moving average, EMA = a × xt + (1 − a) × xt−m , α smoothing factor: 2/(1 + k), k is time period of k day exponential
moving average, LL!t and HHt mean lowest low ! and highest high in the last t days, respectively, Mt : (Ht +Lt +Ct )/3;
Pn Pn  
SMt : Mt−i+1 /n , Dt = Mt−i+1 − SMt  /n, Upt means the upward price change, Dwt means the downward

i=1 i=1
price change at time t.

The input variables utilized in this work are technical indicators described in Table 4 and the
direction of change in the daily Borsa Istanbul (BIST 100) SM index. The entire data contains the period
Entropy 2020, 22, 1239 6 of 19

from 28 March 1996 to 7 February 2020, providing a total of 5968 trading day observations. Furthermore,
information about opening and closing price is available for each trading day. The number of entire
data with decreasing direction is 2827 (47.36%), whereas the number of entire data with increasing
direction is 3141 (52.64%). All the data were obtained from Matriks Information Delivery Services Inc.
(https://www.matriksdata.com).

3.2. Methods

3.2.1. Multilayer Perceptron (MLP)

The architecture of ANN is based on connections of layers by nodes called neurons as well as
the biological neurons of brain [50]. Each path transmits a signal among neurons in a manner similar
to that of synapses [51]. MLP, as a feedforward ANN, contains three main parts: one input layer,
one or2020,
Entropy more 22,hidden
x layers and one output layer, which can be successfully employed for prediction, 7 of 20
classification, signal processing and error filtering [52]. Each node employs one nonlinear function.
MLP
MLP employs
employs backpropagation learning algorithm
backpropagation learning algorithm forfor training
training process
process [53,54].
[53,54]. MLP
MLP asas popular
popular andand
frequently usedtechniques
frequently used techniquesamong
amongother
otherMLPs
MLPs waswas employed
employed to predict
to predict the the direction
direction value.
value. MLPMLP was
was developed
developed by thebyusetheofuse of MATLAB
MATLAB software.
software. Figure Figure 1 indicates
1 indicates the architecture
the architecture of developed
of developed network.
network.
Initially, the network divided data into two sets of training data (with a share of 80%)of
Initially, the network divided data into two sets of training data (with a share 80%)
and and
testing
testing dataa(with
data (with shareaof share
20%) ofrandomly.
20%) randomly.
In theInfirst
thestep
first of
step
theoftraining
the training process,
process, training
training to find
to find the
the optimum number of neurons in the hidden layer. In each training process,
optimum number of neurons in the hidden layer. In each training process, Mean Square Error (MSE) Mean Square Error
(MSE)
was was computed
computed as the performance
as the performance function.
function.

Input layer

SMA
Hidden layer
WMA
MOM
Hidden layer
STOCH
RSI
Tanh(x) Direction
MACD
LWR
A/D
CCI

Figure 1.
Figure The architecture
1. The architecture of
of MLP.
MLP.

Genetic algorithm
Genetic algorithm (GA)(GA) and
and particle
particle swarm
swarm optimization
optimization (PSO),
(PSO), as
as evolutionary
evolutionary algorithms,
algorithms,
have been employed to train the neural network. This approach of hybridization
have been employed to train the neural network. This approach of hybridization of ANN has of ANN has aa lot
lot of
of
advantages such as increasing the accuracy of ANN by updating the weights and
advantages such as increasing the accuracy of ANN by updating the weights and bias values using bias values using
GA and
GA and PSO
PSO [55,56].
[55,56]. The
The aim
aim of
of this
this study
study isis to
to estimate
estimate the
the weights
weights of of hidden
hidden and
and output
output layers
layers of
of
an ANN architecture using GA and PSO during a convergence and accurate
an ANN architecture using GA and PSO during a convergence and accurate estimation process to estimation process to
generate accurate
generate accurate results,
results, and,
and, on
on the
the other
other hand,
hand, toto control
control the
the deviation
deviation from
from target
target point
point in
in such
such aa
way that it prevents deviation and large errors even in different performances. However,
way that it prevents deviation and large errors even in different performances. However, the neural the neural
network needs
network needs to
to be
be left
left alone
alone due
due to
to the
the random
random selection
selection of
of aa sample
sample inin order
order to
to arrive
arrive at
at an
an answer
answer
with
with appropriate
appropriateaccuracy.
accuracy.Therefore,
Therefore,this cancan
this be attributed to the
be attributed to stability and reliability
the stability of the neural
and reliability of the
network through GA and PSO.
neural network through GA and PSO.

3.2.2. Genetic Algorithm (GA)

GA is a subset of approximation techniques in computer science to estimate a proper solution
for optimization problems. GA is a type of evolutionary algorithm (EA) that employs biological
techniques such as heredity and mutations [57,58]. The hypothesis begins with a completely random
population and continues for generations. In each generation, the total population capacity is
assessed, several individuals are selected in a random process from the current generation (based on
Entropy 2020, 22, 1239 7 of 19

3.2.2. Genetic Algorithm (GA)

GA is a subset of approximation techniques in computer science to estimate a proper solution
for optimization problems. GA is a type of evolutionary algorithm (EA) that employs biological
techniques such as heredity and mutations [57,58]. The hypothesis begins with a completely random
population and continues for generations. In each generation, the total population capacity is assessed,
several individuals are selected in a random process from the current generation (based on competencies)
and modified (deducted or re-combined) to form a new generation, and the next repetition of the
algorithm becomes the current generation [59–61].
The optimization process in the genetic algorithm is based on a randomly guided process.
This method is based on Darwin’s theory of gradual evolution and fundamental ideas. In this method,
a set of objective parameters is randomly generated for a fixed number of so-called populations. Or the
fit of that set of information is attributed to that member of that population [62–64]. This process is
repeated for each of the created members, then formed by calling the operators of the genetic algorithm,
including mutation and next-generation selection, and this process will continue until the convergence
criterion is met [59,65]. There are three common criteria for stopping: Algorithm execution time,
Entropy 2020, 22, x 8 of 20
the number of generations that are created and the convergence of the error criterion. The process of
implementing GA, which is the basis of evolutionary algorithms, is presented in Figure 2. which is
of implementing GA, which is the basis of evolutionary algorithms, is presented in Figure 2. which is
adapted and regenerated from [66].
adapted and regenerated from [66].

Choosing
Evaluation
Parents
Parents
Initialization

Evaluation
Population

Evaluation

End
Generation

Figure 2. The
Figure 2. The implementation
implementation process
process of
of GA
GA.

The
The main
main components
components of of the Genetic algorithm include: representation
representation of of the
the environment,
environment,
evaluation
evaluation function,
function, Population
Population (set (set of
of answers),
answers), the the process
process of of choosing
choosing parents,
parents, Operators
Operators of of
Diversity
Diversity (Generation),
(Generation), The The process
process ofof selecting
selecting the
the living
living (choosing
(choosing the the best
best population
population to to build
build the
the
next
next generation)
generation) and and stop condition. Genetic Genetic organization
organization determines
determines how how each
each person
person displays
displays
themselves
themselvesand andbehaves,
behaves,and andtheir
theirphysical
physical quality.
quality.Differences
Differences in genetic
in geneticorganization
organization are are
oneone
of the
of
criteria for for
the criteria distinguishing
distinguishing between
between different
differentmethods
methodsofofevolutionary
evolutionarycomputation.
computation. The The genetic
genetic
algorithm
algorithm uses linearlinear binary
binary organization.
organization.The Themost
moststandard
standardtypetypeofofthis
this organization
organization is the
is the useuse of
of an
an array
array of of bits.
bits. OfOf course,
course, anan arrayofofother
array othertypes
typesofofdata
datacan
canalso
alsobe beused.
used. This
This is due to their constant
constant
size.
size. This
This facilitates
facilitates integration
integration operations
operations [61,67,68].
[61,67,68]. However,
However,itit is is possible
possible toto use
use variable
variable length
length
structures
structures in organizing GA, which makes the implementation of integration very complex.
In this research,
In this research,thethe genetic
genetic algorithm
algorithm was was utilized
utilized to findtothe
find the optimal
optimal point of point
complex of nonlinear
complex
nonlinear
functions in functions in integrating
integrating with the
with the artificial artificial
neural network. neural network.
Genetic Genetic
algorithms algorithms
optimize optimize
artificial neural
artificial neural network
network weights and bias weights
values. Inand fact,bias
the values.
objectiveInfunction
fact, the of objective
the geneticfunction
algorithm of isthe genetic
a function
algorithm is a function
of the statistical of the
results of theMLP.
statistical results
To train, the of the MLP.
P number of To train, theofPeach
population number of population
generation, MLP was of
each generation,
randomly MLPand
initialized was therandomly
error rateinitialized and the
was calculated error
using rate was
training data.calculated using
In the next training
step, data.
the network
In the next step,were
characteristics the network
updatedcharacteristics
according to the were updated
input according
and output to the
values. input
The and output
training processvalues.
of the
The training
algorithm was process of the
repeated until algorithm
the networkwasfeatures
repeated until the taking
improved, network intofeatures
accountimproved,
the newly taking into
population.
account the newly population. In the last step, the output gathered from the network execution was
compared with the actual values and the model execution finished by minimizing the difference
between the two values. Figure 3 presents the flowchart of the MLP–GA algorithm. Table 5 presents
the setting parameters for GA.
Entropy 2020, 22, 1239 8 of 19

In the last step, the output gathered from the network execution was compared with the actual values
and the model execution finished by minimizing the difference between the two values. Figure 3
presents the
Entropy flowchart
2020, 22, x of the MLP–GA algorithm. Table 5 presents the setting parameters for
9 of GA.
20

Input layer

SMA
Hidden layer
WMA
MOM
Hidden layer
STOCH
Direction
RSI
Tanh(x) Fitness function
MACD
LWR
A/D
GA
CCI
Cross-over
mutation

Organized
connection weights

Figure 3. The flowchart of the MLP–GA algorithm.

Figure 3. The flowchart of the MLP–GA algorithm.
3.2.3. Particle Swarm Optimization
Table 5.(PSO)
The characteristics of the MLP–GA.
PSO is a popular and robust optimization method to deal with problems in the n-dimensional
Input Neuron 9
space. The PSO is a mass search algorithm that is modeled on the social behavior of bird groups.
Hidden layer 2
Initially, the algorithm
Hiddenwaslayeremployed
activationfor pattern detection of flight
function of birds at the same time and to
Logsig
suddenly change their path
Output layerand optimize
activation the shape of the handle.
function In PSO,
Gaussian, particles
Tanh (x) flow in the search
space which is affected by Pop.their type
experience and knowledge ofDouble their neighbors;
vector thus the position of
another particle mass affects howsize
Pop. a particle is searched. Results
50,of100
recognition
and 150 of this behavior is the
searching for particlesCrossover
to reach successful
function areas. The particles follow each other and move towards
Scattered
their best neighbors. InCrossover fraction
PSO particles are regulated throughout their0.8 neighborhood [69–72].
At the beginning of Selection function
the work, a group of particles are producedUniform
to reach the best solution. Each
Migration interval
particle is updated through the finest position (pbest) and the finest 10position ever obtained by the
particle population used Migration fraction (gbest) in each step, which
by the algorithm 0.2is presented in Figure 4 based
on an adaptation from [73–77]. Updating the velocity and location of each particle is the next step
3.2.3.after finding
Particle the best
Swarm values, Equations
Optimization (PSO) (1) and (2):
( + 1) = ( ) + 1 × ( )×( ( )− ( )+ 2× ( )×( ( ) (1)
PSO is a popular and robust optimization method to deal with problems in the n-dimensional
space. The PSO is a−mass search( )) algorithm that is modeled on the social behavior of bird groups.
Initially, the algorithm was employed ( for = ( +detection
+ 1)pattern 1) + ( ) of birds at the same time
of flight (2) and to
suddenly change their path and optimize the shape of the handle. In PSO, particles flow in the search
space which is affected by their experience and gbest knowledge of their neighbors; thus the position of

another particle mass affects how a particle is searched. Results of recognition of this behavior is the
searching for particles to reach successful areas. The particles follow each other and move towards
Companionable

their best neighbors. In PSO particles are regulated throughout their neighborhood [69–72].
At the beginning of the work, a group of particles are produced to reach the best solution.
Each particle is updated through the finest position (pbest) and the finest position ever obtained by the
particle population used by the algorithm (gbest) in each step, which is presented in Figure 4 based on
V
an adaptation from [73–77]. Updating the velocity X and location of each particle is the next step after
finding the best values, Equations (1) and (2):

v(t + 1) = v(t) + c1 × rand(pbest

t) × (pbest(t) − position(t) + c2 × rand(t) × ( gbest(t) − position(t)) (1)
New X
position(t + 1) = v(t + 1) + position(t) (2)
Figure 4. The performance of PSO.
on an adaptation from [73–77]. Updating the velocity and location of each particle is the next step
after finding the best values, Equations (1) and (2):
( + 1) = ( ) + 1 × ( )×( ( )− ( )+ 2× ( )×( ( ) (1)

Entropy 2020, 22, 1239

− ( )) 9 of 19
( + 1) = ( + 1) + ( ) (2)

gbest

Companionable
V
X

pbest

New X

Figure4.4. The
Figure The performance
performance of
of PSO.
PSO.

Equation (1) has three parts in the right side, current particle velocity v(t), the second part
c1 × rand(t) × (pbest(t) − position(t)) and third part c2 × rand(t) × ( gbest(t) − position(t)) are responsible
for the rate of change of particle velocity and its direction towards the best personal experience
(nostalgia) and the finest experience of the group (collective intelligence). If the first part is not
considered in this equation v(t), then the velocity of the particles is determined according to the current
position and the best particle experience, and in practice the effect of the current velocity and its inertia
is eliminated. Accordingly, the best particle in the group stays in place, and the others move toward
that particle. In fact, the mass movement of particles without the first part of Equation (1) will be a
process in which the search space gradually shrinks and local search is formed around the best particle.
The parameters c1 and c2 (the value is about 2) determine the importance and weight of collective
intelligence and nostalgia [74–76]. As for the condition of stopping, the following ways are available:

• A certain number of repetitions,

• Achieve a decent threshold,
• A number of repetitions that do not change the competence (for example, if after 10 repetitions
the competency was constant and did not improve),
• The last way is based on the aggregation density around the optimal point.

One of the advantages of PSO over GA is the simplicity and its low parameters. Selecting the
best values for the cognitive and social component leads to accelerating the algorithm and preventing
premature convergence occurs locally at optimal points. In PSO optimization, the proposed variables
are included in the training of a neural network, including network weights and bias. The process is as
follows: First, N vector of position Xi, which N is equal to the number of members of the category, is
generated randomly. The neural network is executed according to the parameters equal to the variables
of these vectors and the error obtained from each run is considered as the degree of fit of the variable
vector of that network. This process is repeated till the final convergence is achieved. The ultimate
convergence is to achieve the optimal position vector (values of optimal weights and bias) so that the
training error is minimized. So the objective function in this optimization need to be minimized as the
amount of error Forecast [78–80]. Table 6 presents the setting values of the MLP–PSO.
Entropy 2020, 22, 1239 10 of 19

Table 6. The characteristics of the MLP–PSO.

Input Neuron 9
Hidden layer 2
Hidden layer activation function logsig
Output layer activation function Gaussian, Tanh (x)
Number of Max. Iteration 500
Pop. size 50, 75, 100 and 125
c1 2
c2 2

3.2.4. Training Phase

Training process categorized into two main steps. First is to select the best architecture of the
ANN, the second is to integrate MLP with optimizers. Therefore, training was developed with 10 to
19 neurons in the first hidden layer and 2 neurons in the second hidden layer with the 80 percent of total
data, according to Table 7. MLP models called as Models 1–6. After this process, MLP was integrated
with GA using population size 50, 100 and 150 (models 7–9, respectively) and PSO using particle size
50, 75, 100 and 125 (models 10–13, respectively). Training was performed into two scenarios, with Tanh
(x) as the output function and with Gaussian function as the default function.

Table 7. The description of the developed models.

Entropy 2020, 22, x 11 of 20

Model 1 MLP (9-10-2-1) Model 8 MLP–GA (100)

Table 7. The description of the developed models.
Model 2 MLP (9-12-2-1) Model 9 MLP–GA (150)
Model 3 Model MLP
1 MLP (9-14-2-1)
(9-10-2-1) Model Model
8 10
MLP–GA (100)MLP–PSO (50)
Model 4 Model MLP
2 MLP (9-15-2-1)
(9-12-2-1) Model Model
9 11
MLP–GA (150)MLP–PSO (75)
Model 5 Model MLP
3 MLP (9-14-2-1) Model
(9-17-2-1) 10 MLP–PSO
Model 12 (50)
MLP–PSO (100)
Model 6 Model MLP
4 MLP (9-15-2-1) Model
(9-19-2-1) 11 MLP–PSO
Model 13 (75)
MLP–PSO (125)
Model 7 Model 5MLP–GA
MLP (9-17-2-1)
(50) Model 12 MLP–PSO (100)
Model 6 MLP (9-19-2-1) Model 13 MLP–PSO (125)
Model 7 MLP–GA (50)
Figure 5 indicates a sample of the training process with MLP–GA which is extracted from the
Figure
training process. The5 indicates a sample
rest of data (20 of the training
percent) wasprocess with MLP–GA
employed which
in testing is extracted from the
process.
training process. The rest of data (20 percent) was employed in testing process.

Figure 5. MLP–GA training process.

Figure 5. MLP–GA training process.
3.2.5. Evaluation Metrics
Table 8 presents evaluation criteria which compares predicted and output values. These
equations are metrics for indicating the performance of models in predicting the target values as the
model output values. In fact, this metrics compares the output of models and target values to
calculate a value for indicating the accuracy of models [51,56,81].

Table 8. Model Evaluation metrics.

Accuracy and Performance Index Description

Entropy 2020, 22, 1239 11 of 19

3.2.5. Evaluation Metrics

Table 8 presents evaluation criteria which compares predicted and output values. These equations
are metrics for indicating the performance of models in predicting the target values as the model
output values. In fact, this metrics compares the output of models and target values to calculate a
value for indicating the accuracy of models [51,56,81].

Table 8. Model Evaluation metrics.

Accuracy and Performance Index Description

P P P
N (XY) − (X) (Y )
Correlation coefficient = √ P 2 P 2
P 2 P N: Number of Data
[N X −( X) ][N X −( XY) 2 ]
N   X: Target value
MAPE (%) = N1
P  Xi −Yi  Y: Output value.
Xi
r i=1

RMSE = 1 P
N (X − Y )2

4. Results
Training was performed by 80% of total data. Results were evaluated in terms of correlation
coefficient, MAPE and RMSE, according to Tables 9 and 10. Table 9 presents results of the training
step without the use of Tanh (x) (using the Gaussian function as default) as the output function and
Table 10 gives results of the training step with Tanh (x) as the output function.

Table 9. Results for training phase with the Gaussian function as default.

Correlation MAPE Processing Correlation MAPE Processing

Model RMSE Model RMSE
Coefficient (%) Time (s) Coefficient (%) Time (s)
Model 1 0.67 0.741035 32.02% 3.82 Model 8 0.694 0.718928 30.57% 8.33
Model 2 0.68 0.733079 31.55% 4.11 Model 9 0.70 0.713458 30.31% 10.82
Model 3 0.676 0.735209 31.52% 4.97 Model 10 0.692 0.721568 30.40% 6.78
Model 4 0.682 0.730448 30.88% 5.11 Model 11 0.689 0.724479 30.89% 7.32
Model 5 0.689 0.723326 30.84% 5.22 Model 12 0.693 0.720478 30.28% 9.02
Model 6 0.693 0.719818 30.59% 5.30 Model 13 0.704 0.708774 29.93% 10.03
Model 7 0.692 0.720763 30.59% 7.22

Table 10. Results for training phase with tanh(x) function.

Correlation MAPE Processing Correlation MAPE Processing

Model RMSE Model RMSE
Coefficient (%) Time (s) Coefficient (%) Time (s)
Model 1 0.684 0.745674 30.12% 3.82 Model 8 0.709 0.72298 28.79% 7.96
Model 2 0.692 0.738467 29.59% 4.11 Model 9 0.716 0.717001 28.69% 9.87
Model 3 0.69 0.739543 29.62% 4.97 Model 10 0.710 0.720822 28.93% 6.22
Model 4 0.698 0.730832 29.09% 5.11 Model 11 0.703 0.728695 29.03% 7.12
Model 5 0.709 0.724664 29.23% 5.22 Model 12 0.708 0.721266 28.48% 8.45
Model 6 0.707 0.724155 28.66% 5.30 Model 13 0.720 0.712372 28.16% 9.23
Model 7 0.708 0.723435 28.78% 7.05

As is clear from Tables 9 and 10, Model 13 provides higher accuracy compared with other models.
It is also clear that using Tanh (x) as the output function of MLP increases the accuracy of the prediction.
According to the results, the hybrid methods increase the processing time (s) compared with those for
the single methods. This was also claimed by Mosavi et al., 2019 [82] and Ardabili et al., 2019 [83].
The main reason can be due to the optimizing process on setting the weights and bias values of the MLP
which consumes more processing time (s). On the other hand, according to the Tables 9 and 10, it is
clear that the GA requires more processing time compared with that of the PSO. More processing time
also can be due to the complexity of the optimizers [84,85]. Using Tanh (x) reduced the processing time.
Entropy 2020, 22, 1239 12 of 19

Testing Results
Tables 11 and 12 present the testing results, respectively, for the Gaussian function and with Tanh
(x). as is clear, in testing process, results are different from those of training process. According to
results of testing process, Model 7 followed by Model 13 for both scenarios, with Tanh (x) and Gaussian
function, provide higher accuracy and lower error compared with other models. It is clear that the
presence of Tanh (x) as the output function increases the accuracy and reduces the error values.

Table 11. Results for testing phase with the Gaussian function as default.

Correlation Correlation
Model MAPE (%) RMSE MAPE (%) RMSE
Coefficient Coefficient
Model 1 0.648 32.63% 0.759687 Model8 0.681 31.00% 0.730846
Model 2 0.661 32.11% 0.748273 Model 9 0.664 31.44% 0.746575
Model 3 0.657 32.07% 0.752376 Model 10 0.680 31.11% 0.731245
Model 4 0.673 31.31% 0.737857 Model 11 0.663 31.89% 0.747604
Model 5 0.663 31.71% 0.747776 Model 12 0.678 30.97% 0.733873
Model 6 0.671 31.24% 0.740842 Model 13 0.677 31.04% 0.735221
Model 7 0.681 30.84% 0.729959

Table 12. Results for testing phase with Tanh (x).

Correlation Correlation
Model MAPE (%) RMSE Model MAPE (%) RMSE
Coefficient Coefficient
Model 1 0.662 30.92% 0.76042 Model 8 0.692 29.10% 0.734701
Model 2 0.673 30.15% 0.751537 Model 9 0.679 29.95% 0.744885
Model 3 0.669 30.13% 0.753864 Model 10 0.694 29.50% 0.73393
Model 4 0.688 29.54% 0.738487 Model 11 0.674 30.10% 0.749981
Model 5 0.670 30.29% 0.74539 Model 12 0.694 29.20% 0.733235
Model 6 0.684 29.48% 0.740869 Model 13 0.694 29.09% 0.732583
Model 7 0.695 29.16% 0.733063

Figure 6 presents the deviation from target values of all models into two scenarios with the
Gaussian function as the default and with Tanh(x) as the output function. According to Figure 6,
it can be concluded that the presence of Tanh (x) as the output function reduces the range of deviation.
Models with a high accuracy have lower deviation values compared to others.
Figure 7 indicates a simple yet essential form of Taylor diagram for the testing process of the
developed models. This diagram is developed according to correlation coefficient and standard
deviation. A point with a lower standard deviation and higher correlation coefficient has higher
accuracy compared with other points. As is clear from Figure 7, Model 13 and Model 7 present higher
accuracy compared with other models.
According to the results, the advantage of the single models is their lower processing time, but the
lowest accuracy can be the most important limitation and disadvantage of the single models compared
with the hybrid ones. This was also claimed by several researches. In the case of using the hybrid
models, the advantages of MLP–PSO such as higher accuracy and lower processing time overtake
the MLP–GA.
 Model 6 0.684 29.48% 0.740869 Model 13 0.694 29.09% 0.732583
Model 7 0.695 29.16% 0.733063

Figure 6 presents the deviation from target values of all models into two scenarios with the
Gaussian function as the default and with Tanh(x) as the output function. According to Figure 6, it
Entropy 2020,can
22,be concluded that the presence of Tanh (x) as the output function reduces the range of deviation.
1239 13 of 19
Models with a high accuracy have lower deviation values compared to others.

(a)

Entropy 2020, 22, x 14 of 20

(b)

(c)

(d)
Figure 6. Deviation from target values for the developed models. (a,d) with the Gaussian function as
Figure 6. Deviation from target values for the developed models. (a,d) with the Gaussian function as
default, (b,c) with the Tanh (x) function.
default, (b,c) with the Tanh (x) function.
Figure 7 indicates a simple yet essential form of Taylor diagram for the testing process of the
developed models. This diagram is developed according to correlation coefficient and standard
deviation. A point with a lower standard deviation and higher correlation coefficient has higher
accuracy compared with other points. As is clear from Figure 7, Model 13 and Model 7 present higher
accuracy compared with other models.
Entropy 2020, 22, 1239 14 of 19
Entropy 2020, 22, x 15 of 20

Figure 7. Taylor diagram for the best response of the models in testing step.
Figure 7. Taylor diagram for the best response of the models in testing step.
5. Conclusions
According
In tomodeling
this paper, the results,
wastheperformed
advantagebyof MLP–GA
the single and
models is their in
MLP–PSO lower
two processing time, but
scenarios including
the lowest accuracy can be the most important limitation and disadvantage of the single
with Tanh (x) and with the Gaussian function as default as the output function in thirteen categories. models
compared with the hybrid ones. This was also claimed by several researches. In the
Research outcomes were evaluated using RMSE and correlation coefficient values to compare thecase of using the
hybrid models,
accuracy the advantages
and performance of MLP–PSO
of the developed suchinas
models higherand
training accuracy
testingand lower
steps. Basedprocessing time
on the results,
overtake the MLP–GA.
using Tanh (x) as the output function improved the accuracy of models significantly. MLP–PSO with
population size 125 followed by MLP–GA with population size 50 provided higher accuracy in the
5. Conclusions
testing step by RMSE 0.732583 and 0.733063, MAPE of 28.16%, 29.09% and correlation coefficient 0.694
In this
and 0.695, paper, modeling
respectively. was performed
As is clear, by MLP–GA
the only advantage and
of the MLP–PSO
single MLP isinitstwo scenarios
lower including
processing time
with
but Tanh
the (x) anddisadvantage
important with the Gaussian
can befunction
claimedas thedefault
lower as the output
accuracy function
compared within thirteen
the hybridcategories.
models.
Research outcomes
According were
to the results, evaluated
using hybrid using RMSEcould
ML method and correlation
successfullycoefficient
improve the values to compare
prediction the
accuracy.
accuracy andMLP–PSO
Accordingly, performancewithoflower
the developed
processingmodels
time and in training and testing
higher accuracy (as steps.
the mainBased on the results,
advantage of the
using
PSO Tanh (x)with
compared as the output
GA) function
overtakes improved In
the MLP–GA. thethis
accuracy
way, theofproblem
models significantly.
statements were MLP–PSO with
successfully
population
covered size
by the 125 followed
solution presentedby MLP–GA with
in the study. The population size 50for
main limitation provided
the future,higher accuracy
is about in the
presenting
testing
and step aby
beating newRMSE
stock0.732583
market andindex 0.733063, MAPE of 28.16%,
using evolutionary methods.29.09%
Theand correlation
future work will coefficient
address
0.694
the and 0.695,
beating respectively.
of stock market, thatAs istheclear, the only
variance advantage
of stock market ofwill
the single MLP is itsaddressed.
be successfully lower processing
Thus,
time
the but the
return important
variance posesdisadvantage
a limitation ofcan thebe claimed
present the lower accuracy compared with the hybrid
research.
models. According to the results, using hybrid ML method could successfully improve the prediction
Author Contributions: Conceptualization, F.E. and A.M.; methodology, F.E. and A.M.; software, A.M.;
accuracy. Accordingly, MLP–PSO with lower processing time and higher accuracy (as the main
validation, S.A. and A.M.; formal analysis, F.E. and A.M.; investigation, F.E.; resources, F.E.; data curation,
advantage of the PSO
F.E.; writing—original draft compared
preparation,with GA)and
F.E, A.M. overtakes the MLP–GA.
S.A.; writing—review and In this A.M.,
editing, way, S.S.B.
the problem
and F.E.;
visualization, S.A.; supervision, F.E., S.S.B. and A.M.; project administration, A.M.;
statements were successfully covered by the solution presented in the study. The main limitationfunding acquisition, A.M.
for
All authors have read and agreed to the published version of the manuscript.
the future, is about presenting and beating a new stock market index using evolutionary methods.
Funding:
The future Wework
acknowledge the financial
will address support
the beating ofofstock
this work by the
market, Hungarian
that State of
the variance andstock
the European Union
market will be
under the EFOP-3.6.1-16-2016-00010 project and the 2017-1.3.1-VKE-2017-00025 project.
successfully addressed. Thus, the return variance poses a limitation of the present research.
Acknowledgments: The support of Alexander von Humboldt foundation is acknowledged.
Author Contributions:
Conflicts of Interest: The Conceptualization, F.E. and
authors declare no conflict A.M.; methodology, F.E. and A.M.; software, A.M.;
of interest.
validation, S.A. and A.M.; formal analysis, F.E. and A.M.; investigation, F.E.; resources, F.E.; data curation, F.E.;
writing—original draft preparation, F.E, A.M. and S.A.; writing—review and editing, A.M., S.S.B. and F.E.;
visualization, S.A.; supervision, F.E., S.S.B. and A.M.; project administration, A.M.; funding acquisition, A.M.
All authors have read and agreed to the published version of the manuscript.

Funding: We acknowledge the financial support of this work by the Hungarian State and the European Union
under the EFOP-3.6.1-16-2016-00010 project and the 2017-1.3.1-VKE-2017-00025 project.

Acknowledgments: The support of Alexander von Humboldt foundation is acknowledged.

Conflicts of Interest: The authors declare no conflict of interest.

Entropy 2020, 22, 1239 15 of 19

Abbreviations
MLP Multilayer perceptron TEPIX Tehran stock exchange price index
MV Moving variance LDA linear discriminant analysis
GA Genetic algorithms BPNN Back propagation neural network
RMSE Root mean square error BSE-SENSEX Bombay Stock Exchange
ANN Artificial neural network TAPI 10 10-day total amount weight stock price index
MOM momentum ICA Independent Component Analysis
Taiwan Stock Exchange Capitalization Weighted
SVM Support vector machine TAIEX
Stock Index
ROC rate of change KOSPI Korea Composite Stock Price Index
BIST 100 Borsa Istanbul 100 index MAPE Mean Absolute Percentage Error
%D stochastic D CSI 300 Capitalization-weighted SM index
CROBEX Zagreb stock index CNX Nifty Standard & Poor’s CNX Nifty stock index
DAX-30 German DAX-30 DJIA Dow Jones Industrial Average
HIS Hang Seng Index FTSE Financial Times Stock Exchange
k-NN k-nearest neighbor QDA quadratic discriminant analysis
S&P 500 Standard & Poor’s 500 GMM Gaussian mixture model
RBF Radial basis function RKELM Robust kernel extreme learning machine
SMA simple moving average KLCI Kuala Lumpur Composite Index
RSI relative strength index BOVESPA Bolsa de Valores de São Paulo
%K stochastic K PNN Probabilistic neural network
%R Larry William’s R% A/D Accumulation/Distribution
OSCP price oscillator CCI Commodity Channel Index
SO Stochastic oscillator MSO Moving stochastic oscillator
SSO Slow stochastic oscillator PSO Particle swarm optimization
MVR Moving variance ratio EMA Exponential moving average
LRL Linear regression line MACD Moving average convergence and divergence
NB Naive Bayes SM Stock market
RS Rough sets DWT Discrete wavelet transform
IBEX-35 Spanish SM ARIMA Autoregressive integrated moving average

References
1. Adebiyi, A.A.; Ayo, C.K.; Adebiyi, M.O.; Otokiti, S.O. Stock price prediction using neural network with
hybridized market indicators. J. Emerg. Trends Comput. Inf. Sci. 2012, 3, 1–9.
2. Hafezi, R.; Shahrabi, J.; Hadavandi, E. A bat-neural network multi-agent system (BNNMAS) for stock price
prediction: Case study of DAX stock price. Appl. Soft Comput. 2015, 29, 196–210. [CrossRef]
3. Liu, G.; Wang, X. A new metric for individual stock trend prediction. Eng. Appl. Artif. Intell. 2019, 82, 1–12.
[CrossRef]
4. Rahman, A.; Saadi, S. Random walk and breaking trend in financial series: An econometric critique of unit
root tests. Rev. Financ. Econ. 2008, 17, 204–212. [CrossRef]
5. Lo, A.W.; MacKinlay, A.C. Stock market prices do not follow random walks: Evidence from a simple
specification test. Rev. Financ. Stud. 1988, 1, 41–66. [CrossRef]
6. Vejendla, A.; Enke, D. Performance evaluation of neural networks and GARCH models for forecasting
volatility and option strike prices in a bull call spread strategy. J. Econ. Policy Res. 2013, 8, 1.
7. Chiang, W.-C.; Enke, D.; Wu, T.; Wang, R. An adaptive stock index trading decision support system. J. Econ.
Policy Res. 2016, 59, 195–207. [CrossRef]
8. Hadavandi, E.; Shavandi, H.; Ghanbari, A. Integration of genetic fuzzy systems and artificial neural networks
for stock price forecasting. Knowl. Based Syst. 2010, 23, 800–808. [CrossRef]
9. Enke, D.; Mehdiyev, N. Stock market prediction using a combination of stepwise regression analysis,
differential evolution-based fuzzy clustering, and a fuzzy inference neural network. Intell. Autom. Soft Comput.
2013, 19, 636–648. [CrossRef]
10. Qiu, M.; Song, Y.; Akagi, F. Application of artificial neural network for the prediction of stock market returns:
The case of the Japanese stock market. Chaos Solitons Fractals 2016, 85, 1–7. [CrossRef]
Entropy 2020, 22, 1239 16 of 19

11. Gorgulho, A.; Neves, R.; Horta, N. Applying a GA kernel on optimizing technical analysis rules for stock
picking and portfolio composition. Expert Syst. Appl. 2011, 38, 14072–14085. [CrossRef]
12. Lin, X.; Yang, Z.; Song, Y. Intelligent stock trading system based on improved technical analysis and Echo
State Network. Expert Syst. Appl. 2011, 38, 11347–11354. [CrossRef]
13. Cervelló-Royo, R.; Guijarro, F.; Michniuk, K. Stock market trading rule based on pattern recognition and
technical analysis: Forecasting the DJIA index with intraday data. Expert Syst. Appl. 2015, 42, 5963–5975.
[CrossRef]
14. Chen, Y.-S.; Cheng, C.-H.; Tsai, W.-L. Modeling fitting-function-based fuzzy time series patterns for evolving
stock index forecasting. Appl. Intell. 2014, 41, 327–347. [CrossRef]
15. Yao, J.; Tan, C.L.; Poh, H.-L. Neural networks for technical analysis: A study on KLCI. Int. J. Theor. Appl. Financ.
1999, 2, 221–241. [CrossRef]
16. Kim, K.-J. Financial time series forecasting using support vector machines. Neurocomputing 2003, 55, 307–319.
[CrossRef]
17. Nair, B.B.; Mohandas, V.; Sakthivel, N.R. A decision tree—Rough set hybrid system for stock market trend
prediction. Int. J. Comput. Appl. 2010, 6, 1–6. [CrossRef]
18. De Oliveira, F.A.; Nobre, C.N.; Zarate, L.E. Applying Artificial Neural Networks to prediction of stock
price and improvement of the directional prediction index—Case study of PETR4, Petrobras, Brazil.
Expert Syst. Appl. 2013, 40, 7596–7606. [CrossRef]
19. Shakeri, B.; Zarandi, M.F.; Tarimoradi, M.; Turksan, I. Fuzzy clustering rule-based expert system for stock
price movement prediction. In Proceedings of the 2015 Annual Conference of the North American Fuzzy
Information Processing Society (NAFIPS), Redmond, WA, USA, 17–19 August 2015; pp. 1–6.
20. Patel, J.; Shah, S.; Thakkar, P.; Kotecha, K. Predicting stock and stock price index movement using trend
deterministic data preparation and machine learning techniques. Expert Syst. Appl. 2015, 42, 259–268.
[CrossRef]
21. Santoso, M.; Sutjiadi, R.; Lim, R. Indonesian Stock Prediction using Support Vector Machine (SVM). In MATEC
Web of Conferences; EDP Sciences: Les Ulis, France, 2018; p. 01031.
22. Chandar, S.K. Stock market prediction using subtractive clustering for a neuro fuzzy hybrid approach.
Clust. Comput. 2019, 22, 13159–13166. [CrossRef]
23. Cervelló-Royo, R.; Guijarro, F. Forecasting stock market trend: A comparison of machine learning algorithms.
Financ. Mark. Valuat. 2020, 6, 37–49.
24. Kara, Y.; Boyacioglu, M.A.; Baykan, Ö.K. Predicting direction of stock price index movement using artificial
neural networks and support vector machines: The sample of the Istanbul Stock Exchange. Expert Syst. Appl.
2011, 38, 5311–5319. [CrossRef]
25. Karymshakov, K.; Abdykaparov, Y. Forecasting stock index movement with artificial neural networks:
The case of istanbul stock exchange. Trak. Univ. J. Soc. Sci. 2012, 14, 231–241.
26. Lahmiri, S. Minute-ahead stock price forecasting based on singular spectrum analysis and support vector
regression. Appl. Math. Comput. 2018, 320, 444–451. [CrossRef]
27. Pulido, M.; Melin, P.; Castillo, O. Particle swarm optimization of ensemble neural networks with fuzzy
aggregation for time series prediction of the Mexican Stock Exchange. Inf. Sci. 2014, 280, 188–204. [CrossRef]
28. Lahmiri, S. Interest rate next-day variation prediction based on hybrid feedforward neural network, particle
swarm optimization, and multiresolution techniques. Phys. A Stat. Mech. Appl. 2016, 444, 388–396.
[CrossRef]
29. Aldin, M.M.; Dehnavi, H.D.; Entezari, S. Evaluating the employment of technical indicators in predicting
stock price index variations using artificial neural networks (case study: Tehran Stock Exchange). Int. J.
Bus. Manag. 2012, 7, 25.
30. Gurjar, M.; Naik, P.; Mujumdar, G.; Vaidya, T. Stock market prediction using ANN. Int. Res. J. Eng.
Technol. (IRJET) 2018, 5, 2758–2761.
31. Yu, L.; Wang, S.; Lai, K.K. Mining stock market tendency using GA-based support vector machines.
In Proceedings of the First International Workshop on Internet and Network Economics, Hong Kong, China,
15–17 December 2005; pp. 336–345.
32. Lu, C.-J. Integrating independent component analysis-based denoising scheme with neural network for
stock price prediction. Expert Syst. Appl. 2010, 37, 7056–7064. [CrossRef]
Entropy 2020, 22, 1239 17 of 19

33. Dash, R.; Dash, P. Efficient stock price prediction using a self evolving recurrent neuro-fuzzy inference system
optimized through a modified differential harmony search technique. Expert Syst. Appl. 2016, 52, 75–90.
[CrossRef]
34. Lahmiri, S. A comparison of PNN and SVM for stock market trend prediction using economic and technical
information. Int. J. Comput. Appl. 2011, 29, 24–30.
35. Dastgir, M.; Enghiad, M.H. Short-term prediction of Tehran Stock Exchange Price Index (TEPIX):
Using artificial neural network (ANN). J. Secur. Exch. 2012, 4, 237–261.
36. Dunis, C.L.; Rosillo, R.; de la Fuente, D.; Pino, R. Applications. Forecasting IBEX-35 moves using support
vector machines. Neural Comput. Appl. 2013, 23, 229–236. [CrossRef]
37. Lahmiri, S.; Boukadoum, M.; Chartier, S. Information fusion and S&P500 trend prediction. In Proceedings of
the 2013 ACS International Conference on Computer Systems and Applications (AICCSA), Ifrane, Morocco,
27–30 May 2013; pp. 1–7.
38. Wang, S.; Shang, W. Forecasting direction of China security index 300 movement with least squares support
vector machine. Procedia Comput. Sci. 2014, 31, 869–874. [CrossRef]
39. Anbalagan, T.; Maheswari, S.U. Classification and prediction of stock market index based on fuzzy metagraph.
Procedia Comput. Sci. 2015, 47, 214–221. [CrossRef]
40. Ballings, M.; Van den Poel, D.; Hespeels, N.; Gryp, R. Evaluating multiple classifiers for stock price direction
prediction. Expert Syst. Appl. 2015, 42, 7046–7056. [CrossRef]
41. Anish, C.; Majhi, B. Hybrid nonlinear adaptive scheme for stock market prediction using feedback FLANN
and factor analysis. J. Korean Stat. Soc. 2016, 45, 64–76. [CrossRef]
42. Jabbarzadeh, A.; Shavvalpour, S.; Khanjarpanah, H.; Dourvash, D. A multiple-criteria approach for forecasting
stock price direction: Nonlinear probability models with application in S&P 500 Index. Int. J. Appl. Eng. Res.
2016, 11, 3870–3878.
43. Jiao, Y.; Jakubowicz, J. Predicting stock movement direction with machine learning: An extensive study on
S&P 500 stocks. In Proceedings of the 2017 IEEE International Conference on Big Data (Big Data), Boston,
MA, USA, 11–14 December 2017; pp. 4705–4713.
44. Garcia, F.; Guijarro, F.; Oliver, J.; Tamošiūnienė, R. Hybrid fuzzy neural network to predict price direction in
the German DAX-30 index. Technol. Econ. Dev. Econ. 2018, 24, 2161–2178. [CrossRef]
45. Nadh, V.L.; Prasad, G.S. Stock market prediction based on machine learning approaches. In Computational
Intelligence and Big Data Analytics; Springer: Berlin/Heidelberg, Germany, 2019; pp. 75–79.
46. Manojlović, T.; Štajduhar, I. Predicting stock market trends using random forests: A sample of the Zagreb stock
exchange. In Proceedings of the 2015 38th International Convention on Information and Communication
Technology, Electronics and Microelectronics (MIPRO), Opatija, Croatia, 25–29 May 2015; pp. 1189–1193.
47. Malagrino, L.S.; Roman, N.T.; Monteiro, A.M. Forecasting stock market index daily direction: A Bayesian
Network approach. Expert Syst. Appl. 2018, 105, 11–22. [CrossRef]
48. Dash, R.; Samal, S.; Rautray, R.; Dash, R. A TOPSIS approach of ranking classifiers for stock index price
movement prediction. In Soft Computing in Data Analytics; Springer: Berlin/Heidelberg, Germany, 2019;
pp. 665–674.
49. Bisoi, R.; Dash, P.; Parida, A. Hybrid Variational Mode Decomposition and evolutionary robust kernel
extreme learning machine for stock price and movement prediction on daily basis. Appl. Soft Comput. 2019,
74, 652–678. [CrossRef]
50. Ecer, F. Artificial Neural Networks in Predicting Financial Performance: An Application for Turkey’s Top
500 Companies. Econ. Comput. Econ. Cybern. Studies Res. 2013, 47, 103–114.
51. Ardabili, S.F.; Mahmoudi, A.; Gundoshmian, T.M. Modeling and simulation controlling system of HVAC
using fuzzy and predictive (radial basis function, RBF) controllers. J. Build. Eng. 2016, 6, 301–308. [CrossRef]
52. Ecer, F. Comparing the bank failure prediction performance of neural networks and support vector machines:
The Turkish case. Econ. Res. 2013, 26, 81–98. [CrossRef]
53. Gundoshmian, T.M.; Ardabili, S.; Mosavi, A.; Varkonyi-Koczy, A.R. Prediction of combine harvester
performance using hybrid machine learning modeling and response surface methodology. In Proceedings of
the International Conference on Global Research and Education, Balatonfüred, Hungary, 4–7 September 2019.
54. Ardabili, S.F. Simulation and Comparison of Control System in Mushroom Growing Rooms Environment.
Master’s Thesis, University of Tabriz, Tabriz, Iran, 2014.
Entropy 2020, 22, 1239 18 of 19

55. Lee, T.K.; Cho, J.H.; Kwon, D.S.; Sohn, S.Y. Global stock market investment strategies based on financial
network indicators using machine learning techniques. Expert Syst. Appl. 2019, 117, 228–242. [CrossRef]
56. Zhong, X.; Enke, D. Predicting the daily return direction of the stock market using hybrid machine learning
algorithms. Financ. Innov. 2019, 1, 14–24. [CrossRef]
57. Harik, G.R.; Lobo, F.G.; Goldberg, D.E. The compact genetic algorithm. IEEE Trans. Evol. Comput. 1999,
3, 287–297. [CrossRef]
58. Khan, W. Predicting stock market trends using machine learning algorithms via public sentiment and
political situation analysis. Soft Comput. 2019, 12, 1–25. [CrossRef]
59. Mühlenbein, H.; Schomisch, M.; Born, J. The parallel genetic algorithm as function optimizer. Parallel Comput.
1991, 17, 619–632. [CrossRef]
60. Kampouropoulos, K.; Andrade, F.; Sala, E.; Espinosa, A.G.; Romeral, L. Multiobjective optimization of
multi-carrier energy system using a combination of ANFIS and genetic algorithms. IEEE Trans. Smart Grid
2018, 9, 2276–2283. [CrossRef]
61. Sahay, R.R.; Srivastava, A. Predicting monsoon floods in rivers embedding wavelet transform, genetic
algorithm and neural network. Water Resour. Manag. 2014, 28, 301–317. [CrossRef]
62. Wang, J.; Wan, W. Optimization of fermentative hydrogen production process using genetic algorithm based
on neural network and response surface methodology. Int. J. Hydrog. Energy 2009, 34, 255–261. [CrossRef]
63. Maertens, K.; De Baerdemaeker, J.; Babuška, R. Genetic polynomial regression as input selection algorithm
for non-linear identification. Soft Comput. 2006, 10, 785–795. [CrossRef]
64. Pandey, P.M.; Thrimurthulu, K.; Reddy, N.V. Optimal part deposition orientation in FDM by using a
multicriteria genetic algorithm. Int. J. Prod. Res. 2004, 42, 4069–4089. [CrossRef]
65. Horn, J.; Nafpliotis, N.; Goldberg, D.E. A niched Pareto genetic algorithm for multiobjective optimization.
In Proceedings of the First IEEE Conference on Evolutionary Computation (IEEE World Congress on
Computational Intelligence), Orlando, FL, USA, 27–29 June 1994; pp. 82–87.
66. Mansoor, A.-K.Z.; Salih, T.A.; Hazim, M.Y. Self-tuning PID Controller using Genetic Algorithm. Iraqi J.
Stat. Sci. 2011, 11, 369–385.
67. Song, S.; Singh, V.P. Frequency analysis of droughts using the Plackett copula and parameter estimation by
genetic algorithm. Stoch. Environ. Res. Risk Assess. 2010, 24, 783–805. [CrossRef]
68. Abdullah, S.S.; Malek, M.; Abdullah, N.S.; Mustapha, A. Feedforward backpropagation, genetic algorithm
approaches for predicting reference evapotranspiration. Sains Malays. 2015, 44, 1053–1059. [CrossRef]
69. Poli, R.; Kennedy, J.; Blackwell, T. Particle swarm optimization. Swarm Intell. 2007, 1, 33–57. [CrossRef]
70. Sun, J.; Feng, B.; Xu, W. Particle swarm optimization with particles having quantum behavior. In Proceedings
of the 2004 Congress on Evolutionary Computation (IEEE Cat. No. 04TH8753), Portland, OR, USA,
19–23 June 2004; pp. 325–331.
71. Parsopoulos, K.E.; Vrahatis, M.N. Particle swarm optimization method in multiobjective problems.
In Proceedings of the 2002 ACM symposium on Applied computing, New York, NY, USA, 22–27 March 2002;
pp. 603–607.
72. Kennedy, J.; Eberhart, R. Particle swarm optimization. In Proceedings of the ICNN’95-International Conference
on Neural Networks, Perth, WA, Australia, 27 November–1 December 1995; pp. 1942–1948.
73. Baltar, A.M.; Fontane, D.G. Use of multiobjective particle swarm optimization in water resources management.
J. Water Resour. Plan. Manag. 2008, 134, 257–265. [CrossRef]
74. Clerc, M. Particle Swarm Optimization; John Wiley & Sons: Hoboken, NJ, USA, 2010; Volume 93.
75. Shi, Y. Particle swarm optimization: Developments, applications and resources. In Proceedings of the 2001
Congress on Evolutionary Computation (IEEE Cat. No. 01TH8546), Seoul, Korea, 27–30 May 2001; pp. 81–86.
76. Shi, Y.; Eberhart, R.C. Parameter selection in particle swarm optimization. In Proceedings of the International
Conference on Evolutionary Programming, San Diego, CA, USA, 25–27 March 1998; pp. 591–600.
77. Allaoua, B.; Laoufi, A.; Gasbaoui, B.; Abderrahmani, A. Technologies. Neuro-fuzzy DC motor speed control
using particle swarm optimization. Leonardo Electron. J. Pract. Technol. 2009, 15, 1–18.
78. Gharghan, S.K.; Nordin, R.; Ismail, M.; Ali, J.A. Accurate wireless sensor localization technique based
on hybrid PSO-ANN algorithm for indoor and outdoor track cycling. IEEE Sensors J. 2015, 16, 529–541.
[CrossRef]
Entropy 2020, 22, 1239 19 of 19

79. Armaghani, D.J.; Raja, R.S.N.S.B.; Faizi, K.; Rashid, A.S.A. Developing a hybrid PSO–ANN model for
estimating the ultimate bearing capacity of rock-socketed piles. Neural Comput. Appl. 2017, 28, 391–405.
[CrossRef]
80. Liou, S.-W.; Wang, C.-M.; Huang, Y.-F. Integrative Discovery of Multifaceted Sequence Patterns by
Frame-Relayed Search and Hybrid PSO-ANN. J. UCS 2009, 15, 742–764.
81. Pang, X.; Zhou, Y.; Wang, P.; Lin, W.; Chang, V. An innovative neural network approach for stock market
prediction. J. Supercomput. 2020, 76, 2098–2118. [CrossRef]
82. Mosavi, A.; Salimi, M.; Faizollahzadeh Ardabili, S.; Rabczuk, T.; Shamshirband, S.; Varkonyi-Koczy, A.R.
State of the art of machine learning models in energy systems, a systematic review. Energies 2019, 12, 1301.
[CrossRef]
83. Liu, H.; Long, Z. An improved deep learning model for predicting stock market price time series.
Digit. Signal Process. 2020, 7, 102741. [CrossRef]
84. Nabipour, M.; Nayyeri, P.; Jabani, H.; Shahab, S.; Mosavi, A. Predicting Stock Market Trends Using Machine
Learning and Deep Learning Algorithms Via Continuous and Binary Data: A Comparative Analysis.
IEEE Access 2020, 8, 150199–150212. [CrossRef]
85. Nabipour, M.; Nayyeri, P.; Jabani, H.; Mosavi, A.; Salwana, E. Deep learning for stock market prediction.
Entropy 2020, 22, 840. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional
affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).